{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:P42XLZOB6RLJ2G7GSHYN375FG2","short_pith_number":"pith:P42XLZOB","canonical_record":{"source":{"id":"2503.01840","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-03T18:59:04Z","cross_cats_sorted":[],"title_canon_sha256":"ed257265660a32c954a9acae0b22dea0f1f57b9ae342a852e7d66a928032e253","abstract_canon_sha256":"d96088f2e47dac06a0977b5581508b768372ecfe4e76df8f9b2ab36b7028e971"},"schema_version":"1.0"},"canonical_sha256":"7f3575e5c1f4569d1be691f0ddffa53688acad4a13f73e1dd00d7f19ef719af9","source":{"kind":"arxiv","id":"2503.01840","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2503.01840","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"arxiv_version","alias_value":"2503.01840v3","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01840","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"pith_short_12","alias_value":"P42XLZOB6RLJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"P42XLZOB6RLJ2G7G","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"P42XLZOB","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:P42XLZOB6RLJ2G7GSHYN375FG2","target":"record","payload":{"canonical_record":{"source":{"id":"2503.01840","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-03T18:59:04Z","cross_cats_sorted":[],"title_canon_sha256":"ed257265660a32c954a9acae0b22dea0f1f57b9ae342a852e7d66a928032e253","abstract_canon_sha256":"d96088f2e47dac06a0977b5581508b768372ecfe4e76df8f9b2ab36b7028e971"},"schema_version":"1.0"},"canonical_sha256":"7f3575e5c1f4569d1be691f0ddffa53688acad4a13f73e1dd00d7f19ef719af9","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:50.539795Z","signature_b64":"BI7Do+xY1kN7uhsdmx0Pjwuym3YBCHVMBAjq3UvlG8hWxW1pwYpKuD11dByz2A8c1Z42EC4wyWN4XHYr4CP3BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7f3575e5c1f4569d1be691f0ddffa53688acad4a13f73e1dd00d7f19ef719af9","last_reissued_at":"2026-05-17T23:38:50.539247Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:50.539247Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2503.01840","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wFaqfiMbFhce2usD1n0ybMSAOaoe/ZSguxgDH86fHxzUvGa6Ckz10cJikeaKL/HyzENmfwVRMdvV2CMOgY9bCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T06:54:55.034156Z"},"content_sha256":"68c1b559c3acb3cacfdc1476ef78ee49465cd40d798b75911f9c4620551626a1","schema_version":"1.0","event_id":"sha256:68c1b559c3acb3cacfdc1476ef78ee49465cd40d798b75911f9c4620551626a1"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:P42XLZOB6RLJ2G7GSHYN375FG2","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Yuhui Li","submitted_at":"2025-03-03T18:59:04Z","abstract_excerpt":"The sequential nature of modern LLMs makes them expensive and slow, and speculative sampling has proven to be an effective solution to this problem. Methods like EAGLE perform autoregression at the feature level, reusing top-layer features from the target model to achieve better results than vanilla speculative sampling. A growing trend in the LLM community is scaling up training data to improve model intelligence without increasing inference costs. However, we observe that scaling up data provides limited improvements for EAGLE. We identify that this limitation arises from EAGLE's feature pre"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"EAGLE-3 abandons feature prediction in favor of direct token prediction and replaces reliance on top-layer features with multi-layer feature fusion via training-time test, significantly enhancing performance and enabling the draft model to fully benefit from scaling up training data, achieving up to 6.5x speedup and 1.4x over EAGLE-2.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That direct token prediction combined with multi-layer feature fusion via training-time test will remove the constraints that previously limited gains from scaling training data, without introducing new accuracy or stability issues in the draft model.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EAGLE-3 reaches up to 6.5x LLM inference speedup by replacing feature prediction with direct token prediction and multi-layer fusion through training-time test.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6cf639fb80120723be5de163add5a8aba12193891a4d2686572c5e14828ee6e6"},"source":{"id":"2503.01840","kind":"arxiv","version":3},"verdict":{"id":"f4dfb1d5-6f6f-44b5-abae-d7de670c8a0a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T18:48:28.987599Z","strongest_claim":"EAGLE-3 abandons feature prediction in favor of direct token prediction and replaces reliance on top-layer features with multi-layer feature fusion via training-time test, significantly enhancing performance and enabling the draft model to fully benefit from scaling up training data, achieving up to 6.5x speedup and 1.4x over EAGLE-2.","one_line_summary":"EAGLE-3 reaches up to 6.5x LLM inference speedup by replacing feature prediction with direct token prediction and multi-layer fusion through training-time test.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That direct token prediction combined with multi-layer feature fusion via training-time test will remove the constraints that previously limited gains from scaling training data, without introducing new accuracy or stability issues in the draft model.","pith_extraction_headline":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"eea725488f826c92de03a8fb0152f69ddd473d002b240d7757c4797b32bd3efd"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"f4dfb1d5-6f6f-44b5-abae-d7de670c8a0a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"xZ4yEAzFdcIfDU1RBnC4eRCh8xmxjWvD6EaGD+RfTDICMW4XN9cy700mBsCY4ALwFNSKoA+rnDin06Ebc0GJBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T06:54:55.034624Z"},"content_sha256":"075db068d57ef7fc743979ed9a73ec2dfe5e79ccfe37bcb4c3161c986f345b57","schema_version":"1.0","event_id":"sha256:075db068d57ef7fc743979ed9a73ec2dfe5e79ccfe37bcb4c3161c986f345b57"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/P42XLZOB6RLJ2G7GSHYN375FG2/bundle.json","state_url":"https://pith.science/pith/P42XLZOB6RLJ2G7GSHYN375FG2/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/P42XLZOB6RLJ2G7GSHYN375FG2/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-30T06:54:55Z","links":{"resolver":"https://pith.science/pith/P42XLZOB6RLJ2G7GSHYN375FG2","bundle":"https://pith.science/pith/P42XLZOB6RLJ2G7GSHYN375FG2/bundle.json","state":"https://pith.science/pith/P42XLZOB6RLJ2G7GSHYN375FG2/state.json","well_known_bundle":"https://pith.science/.well-known/pith/P42XLZOB6RLJ2G7GSHYN375FG2/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:P42XLZOB6RLJ2G7GSHYN375FG2","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"d96088f2e47dac06a0977b5581508b768372ecfe4e76df8f9b2ab36b7028e971","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-03T18:59:04Z","title_canon_sha256":"ed257265660a32c954a9acae0b22dea0f1f57b9ae342a852e7d66a928032e253"},"schema_version":"1.0","source":{"id":"2503.01840","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2503.01840","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"arxiv_version","alias_value":"2503.01840v3","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01840","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"pith_short_12","alias_value":"P42XLZOB6RLJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"P42XLZOB6RLJ2G7G","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"P42XLZOB","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:075db068d57ef7fc743979ed9a73ec2dfe5e79ccfe37bcb4c3161c986f345b57","target":"graph","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"EAGLE-3 abandons feature prediction in favor of direct token prediction and replaces reliance on top-layer features with multi-layer feature fusion via training-time test, significantly enhancing performance and enabling the draft model to fully benefit from scaling up training data, achieving up to 6.5x speedup and 1.4x over EAGLE-2."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That direct token prediction combined with multi-layer feature fusion via training-time test will remove the constraints that previously limited gains from scaling training data, without introducing new accuracy or stability issues in the draft model."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"EAGLE-3 reaches up to 6.5x LLM inference speedup by replacing feature prediction with direct token prediction and multi-layer fusion through training-time test."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference."}],"snapshot_sha256":"6cf639fb80120723be5de163add5a8aba12193891a4d2686572c5e14828ee6e6"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"eea725488f826c92de03a8fb0152f69ddd473d002b240d7757c4797b32bd3efd"},"paper":{"abstract_excerpt":"The sequential nature of modern LLMs makes them expensive and slow, and speculative sampling has proven to be an effective solution to this problem. Methods like EAGLE perform autoregression at the feature level, reusing top-layer features from the target model to achieve better results than vanilla speculative sampling. A growing trend in the LLM community is scaling up training data to improve model intelligence without increasing inference costs. However, we observe that scaling up data provides limited improvements for EAGLE. We identify that this limitation arises from EAGLE's feature pre","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Yuhui Li","cross_cats":[],"headline":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-03T18:59:04Z","title":"EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.01840","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T18:48:28.987599Z","id":"f4dfb1d5-6f6f-44b5-abae-d7de670c8a0a","model_set":{"reader":"grok-4.3"},"one_line_summary":"EAGLE-3 reaches up to 6.5x LLM inference speedup by replacing feature prediction with direct token prediction and multi-layer fusion through training-time test.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"By switching to direct token prediction and multi-layer feature fusion, EAGLE-3 enables draft models to improve with increased training data for faster LLM inference.","strongest_claim":"EAGLE-3 abandons feature prediction in favor of direct token prediction and replaces reliance on top-layer features with multi-layer feature fusion via training-time test, significantly enhancing performance and enabling the draft model to fully benefit from scaling up training data, achieving up to 6.5x speedup and 1.4x over EAGLE-2.","weakest_assumption":"That direct token prediction combined with multi-layer feature fusion via training-time test will remove the constraints that previously limited gains from scaling training data, without introducing new accuracy or stability issues in the draft model."}},"verdict_id":"f4dfb1d5-6f6f-44b5-abae-d7de670c8a0a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:68c1b559c3acb3cacfdc1476ef78ee49465cd40d798b75911f9c4620551626a1","target":"record","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"d96088f2e47dac06a0977b5581508b768372ecfe4e76df8f9b2ab36b7028e971","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-03T18:59:04Z","title_canon_sha256":"ed257265660a32c954a9acae0b22dea0f1f57b9ae342a852e7d66a928032e253"},"schema_version":"1.0","source":{"id":"2503.01840","kind":"arxiv","version":3}},"canonical_sha256":"7f3575e5c1f4569d1be691f0ddffa53688acad4a13f73e1dd00d7f19ef719af9","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"7f3575e5c1f4569d1be691f0ddffa53688acad4a13f73e1dd00d7f19ef719af9","first_computed_at":"2026-05-17T23:38:50.539247Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:50.539247Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"BI7Do+xY1kN7uhsdmx0Pjwuym3YBCHVMBAjq3UvlG8hWxW1pwYpKuD11dByz2A8c1Z42EC4wyWN4XHYr4CP3BQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:50.539795Z","signed_message":"canonical_sha256_bytes"},"source_id":"2503.01840","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:68c1b559c3acb3cacfdc1476ef78ee49465cd40d798b75911f9c4620551626a1","sha256:075db068d57ef7fc743979ed9a73ec2dfe5e79ccfe37bcb4c3161c986f345b57"],"state_sha256":"b1ec18309aab218592c13d91d0de9cc9c2d98124aeefb3771d59ac4226a55e84"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ft21gvt1cCeFJ9Rq0lAuZkz8UdVu23vcraSLCp7gI8QBD6G4kEGTy1CGzmH79oOAbqcSpvBe6uhXWa6XC2ebCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-30T06:54:55.037350Z","bundle_sha256":"0059666f066bcbaa41d27e6b6543aa4a02b0d9342aa23bb91de4adebcd9db6e9"}}